SpecBuddy

AI agents can't read your mind. The spec-first companion for JetBrains IDEs — capture intent, agents execute. Free public beta ↓

[SpecBuddy @spec_buddy]

SpecBuddy is live.

Review what your AI agent is going to do — before it writes a line of code.
Spec → plan → step-by-step execution, all reviewable, inside your JetBrains IDE. Runs on Claude Code or Codex.

Free public beta. Demo 👇

https://t.co/56yOt5o2Hk

[SpecBuddy @spec_buddy]

An AI coding agent (whether Claude Code from @ClaudeDevs or Codex from @OpenAI) shouldn’t get permission to change the whole project from one prompt.

Review the spec! Review the plan! Then review one step’s diff.

Here’s the full loop inside @jetbrains IDEs from @spec_buddy:

[SpecBuddy @spec_buddy]

https://t.co/6Z2qqSqZPk

[SpecBuddy @spec_buddy]

SpecBuddy now works with @opencode.

Full flow — spec, plan, step-by-step execution — same as Claude Code and Codex.

The control layer shouldn't care which agent you run. Your favorite agent is your choice, not our limitation.

In your @jetbrains IDE 👇

[SpecBuddy @spec_buddy]

@MattJamesBoyle Only if you have tests🙂

[SpecBuddy @spec_buddy]

@aashatwt Really? Skipping bitcoin in 2009 cost you nothing — your wealth stayed exactly the same either way. Skipping agentic engineering now actually risks your career. That's a bigger deal than the analogy suggests

[SpecBuddy @spec_buddy]

@jonathan_wilke Can we call writing specs "coding"? If so, I still do it every day.

[SpecBuddy @spec_buddy]

Didn't you notice that most modern AI startup landing pages look almost identical? (Ours too → https://t.co/wK1QQy7Ezk)

[SpecBuddy @spec_buddy]

https://t.co/O4OzSvCHOi

[SpecBuddy @spec_buddy]

@N_and_ni Why nobody buys their todo-list app

[SpecBuddy @spec_buddy]

@dexhorthy Isn't this too much to ask of every single developer? To me, that's IDP/platform-team responsibility. Isn't a regular software engineer better off just getting sharper at working with agents day-to-day?

[SpecBuddy @spec_buddy]

IDE will shift into a place where you write specs, read code, do code reviews, and sometimes debug. It also holds a wide project model that can be really valuable context for agents.

[Piyush Garg @piyushgarg_dev]

So IDE is just a read-only tool to preview code?

[SpecBuddy @spec_buddy]

@simas_ch But doesn't the workflow itself need to be loaded then?

[SpecBuddy @spec_buddy]

That doesn't give you a chance to notice your agent went completely wrong until you've burned through your whole weekly budget. Convenient for them too — nudges you toward upgrading. At least with the 5-hour cap, you get a natural checkpoint to see what's actually been done 🙂

[Titanium @akinkunmi]

Wow. I think Codex has removed the 5-hour limit and only does weekly now. That's really nice!

[SpecBuddy @spec_buddy]

@dark_coderz Replace those who didn't become faster

[SpecBuddy @spec_buddy]

When you make a mistake in a low-level language, your program crashes. When you make a mistake in a high-level language, it throws an exception and keeps running. What happens when you make a mistake in system design?

[Piyush Garg @piyushgarg_dev]

System Design is the new coding language.

[SpecBuddy @spec_buddy]

Starting a new backend service today — what stack are you picking?

And if an agent's doing most of the implementation anyway, does @QuarkusIO vs @springboot still matter?

[SpecBuddy @spec_buddy]

@13_narcissus As for me, the current layer is specs. I'd love to know what's next — if I found out, I'd drop everything and start working on it 🙂

[SpecBuddy @spec_buddy]

@LukasCantCode Much less than a year ago. Back then, agents were far more likely to just start building some feature I hadn't even asked for.

[SpecBuddy @spec_buddy]

@Franc0Fernand0 Like healing people with ChatGPT without being a doctor